@std/async/unstable-semaphore should be bind methods in the constructor #7195

Activity

1. SnowSquire

   added

   bugSomething isn't workingSomething isn't working

   needs triage

   on Jun 26, 2026
2. truffle-dev commented on Jun 29, 2026

   

   truffle-dev

   on Jun 29, 2026

   Contributor

   More actions

   The throw is deterministic, not flaky. release reads this.#head on its first line, and .finally(sema.release) invokes it with this === undefined, so it always throws TypeError: Cannot read properties of undefined (reading '#head').

   What varies is the consequence, which is what made it look inconsistent:

   • The rejection is unobserved (no .catch), so whether you actually see "Uncaught (in promise)" depends on whether the runtime surfaces the unhandled rejection before the program exits.
   • Every detached release throws before restoring the permit, so the permit leaks. Starting from max, each leaked permit shrinks the live pool, and a later acquire only blocks once cumulative leaks exhaust the headroom — so a higher max masks it longer, exactly as you saw.

   Permits never recover:

   const s = new Semaphore(2);
   await s.acquire().then(() => {}).finally(s.release).catch(() => {});
   await s.acquire().then(() => {}).finally(s.release).catch(() => {});
   // both permits leaked; the next acquire() blocks forever

   On the fix: binding in the constructor works, but there's no *.bind(this) in a constructor anywhere in std today. An arrow class field (release = (): void => { ... }) auto-binds, or release stays a prototype method and the doc notes it must be called on the instance. Since this is a concurrency primitive the per-instance allocation tradeoff is a real one. Happy to send a PR for whichever shape you'd prefer.
3. sanjibani

   added a commit that references this issue on Sep 13, 2026

   fix(async/unstable-semaphore): bind methods in constructor (denoland#…

   ...

   d9d2f8b